[
https://issues.apache.org/jira/browse/FLUME-1191?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13469716#comment-13469716
]
Hari Shreedharan commented on FLUME-1191:
-----------------------------------------
You could achieve the same using LoadBalancingSinkSelector. so agent tier will
write randomly or round-robin to the collectors - so the load is not on one
collector.
> Would like autoE2EChain to have the ability to handle different classes of
> agents
> ---------------------------------------------------------------------------------
>
> Key: FLUME-1191
> URL: https://issues.apache.org/jira/browse/FLUME-1191
> Project: Flume
> Issue Type: New Feature
> Components: Sinks+Sources
> Affects Versions: v0.9.4
> Environment: CentOS
> Flume build:
> Flume 0.9.4-cdh3u3
> Git repository https://github.com/cloudera/flume/flume-core
> rev unknown
> Compiled by jenkins on 20120126-1114
> Reporter: Thomas Andrews
>
> We have an agent running on each of our servers, and our servers are writing
> audit data to these agents. The agents are configured to use autoE2EChain
> for their sinks.
> The problem is, out of 60+ servers, five of them generate the vast majority
> of the traffic, and, with this many agents configured to write to our two
> collectors, this means that there is a high probability (about one in three)
> ofautoE2EChain assigning four or five of those servers to write to one
> collector.
> Ideally, we'd have a way to set the sink as: autoE2EChain("high-traffic") so
> that some 'class' of agents would be assigned to collectors separately.
> Alternatively, I suppose, autoE2EChain could pass a weight in:
> autoE2EChain(20).
> I realize most work on the NG Flume, so perhaps the terminology will be
> different there, but the idea is likely still applicable.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
